🌸🔊 Pollinations Audio Gen (BYOP)

Want a voiceover for that video you just rendered, without babysitting an 11GB TTS model or paying for a separate ElevenLabs subscription? That's the pitch for 🌸🔊 Pollinations Audio Gen (BYOP): you type text, pick a model, and it hands you a URL pointing at a freshly generated audio file on the Pollinations cloud. No local GPU, no downloads, no VRAM - the heavy lifting happens on their end. On an 8GB laptop this is the difference between "maybe I'll add a voiceover later" and actually doing it.

This node is one of five in the ComfyUI-Pollinations-BYOP pack (image, video, text, audio, and a login node). They all share the same DNA: build a URL against the Pollinations API, return a string. Learning one teaches you the family.

How it works

Read the source and the "magic" evaporates. The node does two things: resolves your API key (node field first, then the pack's saved config, then the POLLINATIONS_API_KEY env var), then constructs a URL like:

https://gen.pollinations.ai/audio/<url-encoded text>?model=elevenlabs&voice=sarah&key=sk_...

That's the whole mechanism. The return is a STRING called audio_url - and here's the part that trips people up: the node does not download anything. It returns the URL, and the audio is only generated when something actually fetches that URL. So wire audio_url into whatever you use for grabbing remote files (or a node that can open a URL), or just right-click → open in browser to audition it. Don't stare at the node waiting for a file to appear in your output folder - that's not how this one rolls.

The inputs that matter

• text - your script, in the multiline box. TTS models read it literally, so punctuation and phrasing matter.
• model - a dropdown pulled from the pack's models.json (default elevenlabs). The list splits into camps: TTS (elevenlabs, elevenflash, qwen-tts, kokoro), music (elevenmusic, stable-audio-3-large/medium), sound effects (eleven-sfx), and transcription (whisper, universal-2, grok-transcribe, scribe). Anything with a 💎 next to it is paid-only - you'll need a key.
• voice - nine OpenAI-style names (alloy, echo, fable, onyx, nova, shimmer, sarah, rachel, charlie, default sarah). Only TTS models actually use it; music and transcription models ignore it.
• api_key (optional) - a per-node override. Leave blank to use the pack-wide key (set via the BYOP Login node, the Settings menu, or the env var).

One output: audio_url, a string you feed to a downloader or preview.

Installing it

This is a pure API node - no model files, no heavy dependencies, nothing extra for ComfyUI to load. Either use ComfyUI Manager (search "Pollinations BYOP") or:

cd ComfyUI/custom_nodes
git clone https://github.com/ChunkyPanda29/ComfyUI-Pollinations-BYOP.git

Then restart ComfyUI. Your install size doesn't move an inch.

Gotchas

The anonymous tier is free but heavily rate-limited - expect slow generations and timeouts during peak traffic. The README's own answer to "why does my generation fail?" is: add your free API key. It's free, it just requires a login, and the BYOP Login node in this same pack does that login for you. Also remember the 💎 models won't work keyless, and if you're transcription-then-something, the flow is: run the node, fetch the URL, and read the transcript from whatever saves the response.